With the increasing demand of modern society for emergency investigation, public safety monitoring and intelligent data analysis, video monitoring systems are widely applied in society. In recent years, “smart city” projects to promote safe city have started in many big cities in China, e.g., traffic monitoring, campus monitoring, building monitoring, court monitoring, bank monitoring and venue monitoring. Video monitoring can be seen almost everywhere in modern life for people's safety, traffic safety, sanitary safety, societal safety and production safety.
A large amount of monitoring video data usually needs huge storage cost. To reduce storage cost, the conventional technology often sacrifices video quality, which goes against the fundamental objective of video monitoring—capability of identification. Therefore, facing such a large volume of monitoring video data, a video encoding technique with higher compression ratio is needed to save storage cost.
Monitoring video data has its own characteristic. Background scenario information is relatively stable. The foreground and the background of an image can be segmented utilizing a background segregation technology for intelligent analysis and relevant information is sent to an encoder. Then, the encoder can further improve the compression ratio of the background information without influencing quality, thereby reducing the storage cost of video.
The foreground mainly includes targets of interest, moving objects and objects that approach camera. The background mainly refers to targets of no interest, still objects and objects that move away from the camera.
Video cameras for video monitoring include two types, namely, stationary cameras and dynamic cameras. After a stationary camera is installed, the background can be stably obtained. If results of the background-foreground segregation are used for encoding, the bit rate can be reduced without sacrificing encoding quality of target of interest. If a dynamic camera rotates, the background changes accordingly. Maintaining the background may be difficult, and clean background may not be obtained. During the rotation process, foreground and background are used for encoding; therefore the bit rate can increase. Moreover, the reference background can expire, and the encoding quality may suffer, so that semi-structured encoding may not be achieved.
Hence it is highly desirable to improve the techniques for video encoding.